Visualizing sensitivity
String music


Feed You can follow this conversation by subscribing to the comment feed for this post.

Mike Anderson

Famously, he [Wolfgang Pauli] once said of such an unclear paper: "It is not even wrong."

Jon Peltier

An attractive graphic, but a horrible chart.


Can you elaborate? I feel like it's intelligible (as well as graphically interesting). What makes this such a bad chart?


Can you elaborate? I feel like it's intelligible (as well as graphically interesting). What makes this such a bad chart? Merely the skewed perspective?


I agree, I find it reasonably interesting and easy to read. It is sparse, in the sense that 20 data points are given so much space, but it works for me.


Graphs (and tables) are usually meant to invite comparison and then be able to deliver on the invitation! This one doesn't.

1. It needs a legend to show the ranking for each attribute (revenue & fees generated)
2. The angles used prevent any such visual ranking comparison.
3. Ranking assistance also needs to be provided through posting actual values for the revenues & fees generated attributes.
4. The distorted view of the graph adds no visual benefit beyond what a plain old table would show.


The problem with the graphic is that, while getting the individual information is easy, getting the corrilation trend is hard. A simple scatterplot would be much better. (If somewhat boring.)

Matt Alcock

Looks great, but how long did it actually take you to understand what is was trying to visually represent. If this is above the norm then its got to be a junk chart, which this most certainly is? Graphically interesting doesn't always mean visually informative.


In judging these graphics, my first instinct is to ask what questions users might want to answer. If the question is the correlation between revenue rank and fee rank, then this graphic is barely adequate. A simple profile chart will expose that information more clearly.

If the question is the correlation between revenues and fees (since ranks remove scaling information), then this is a terrible chart. Distances don't mean anything here since the expected Cartesian grid is not used. It also fails the sufficiency test.

Finally, we might note that the decimal point is superfluous.

Jon Peltier


My eyes are drawn toward the diagonal baseline, where the two charts meet.

I can't simply look at the bars and intuit the ranks. The dots on the diagonal black lines look like another pair of data series, and the ranks used to label these pseudo-series force me to think too much.

In fact, it takes too much analysis of the chart before one can begin an analysis of its information.

It would have been much more effective as a two-series bar chart or dot plot.

The comments to this entry are closed.